Indices · 2026-10-14
The numbers to watch.
Daily market indices for LLM usage: reasoning share, free traffic, tool-capable and multimodal models, context length, open weights, geography, concentration, and blended price. Formulas in the methodology.
462 models catalogued upstream on OpenRouter as of 2026-10-15.
Reasoning share
91.52%
−0.69pp · 30d
Free-traffic share
6.37%
−0.49pp · 30d
HHI (top-10)
651
+61.04 pts · 30d
Median $ / M paid
$0.269
−$0.181 · 30d
Tool-capable share
97.28%
−2.24pp · 30d
Context 100K+
62.36%
+7.96pp · 30d
Image-capable share
75.06%
+16.41pp · 30d
Top 50 coverage
93.56%
−0.03pp · 30d
Top 50 models cover 93.6% of tokens on 2026-10-14, measured exactly from OpenRouter's daily total for every model outside the top 50.
Reasoning Share Index
Share of tracked tokens going to reasoning-flagged models.
Free-Traffic Index
Share of tokens routed through :free variants.
Tool-Calling Share (model capability)
Share of all tokens served by models that support tool calling. OpenRouter filters by model capability, not by whether each request called a tool.
Tool calling
97.3% of OpenRouter tokens went to models that support tool calling on 2026-10-14. Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-14. tidelines.ai/indices
Context-length mix
Share of tokens by request context size. Buckets sum to 100%.
Long context
Requests with 100K+ tokens of context made up 62.4% of OpenRouter tokens on 2026-10-14. Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-14. tidelines.ai/indices
Modality mix (model capability)
Share of all tokens served by models that accept image input, produce image output, or handle audio. OpenRouter filters these by model capability, not by request.
Modality
On 2026-10-14, 75.1% of OpenRouter tokens went to models that accept image input, 0.02% to models that generate images and 11.28% to audio-capable models. Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-14. tidelines.ai/indices
Code vs natural language
Weekly estimate (sampled). Share of classified tokens written in programming languages versus natural language.
Code vs natural language · weekly estimate (sampled)
In the week of 2026-10-12, an estimated 24.8% of classified OpenRouter tokens were code (weekly sampled estimate). Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-14. tidelines.ai/indices
HHI top-10 concentration
Herfindahl-Hirschman on the daily top-10 model shares. Higher = more concentrated.
Median blended price paid
Usage-weighted median of (0.3·prompt + 0.7·completion) $ / M tokens across all tracked models.
Peak tracked volume: 27T tokens/day.
Open weights vs closed
Share of tracked tokens routed to open-weights models.
China vs Western labs
Share of tracked tokens routed to Chinese labs versus US & European labs.
Top-10 HHI
Top-10 model share HHI stood at 651 on 2026-10-14 — higher = more concentrated. Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-14. tidelines.ai/indices